I’m using jboss and cmt and have seen strange behaviour when using requires_new on an ejb method that I loop over from another bean to insert some records. I see that intermittently after restarting the jboss the operations in the method aren’t committed to db after the method is finished.So, I have tried to use TransactionSynchronizationRegistry and found that when the data isn’t committed I actually don’t get a new transaction each time the method is entered. Anyone who has heard of jboss acting this way? I'm using ejb, jboss, jpa, Hibernate, cmt.
So, I've finally solved this problem. I used this great blog post
http://piotrnowicki.com/2011/11/am-i-in-the-same-transaction-am-i-using-the-same-persistencecontext/
to make sure that I actually was in the same transaction after entering the method annotated REQUIRES_NEW. So then I understood that JBoss intermittently ignored the annotation and defaulted to REQUIRED instead. This was because multiple classes implemented the same interface. So I simplified it so that I now have one interface, annotated @Local, and one implementing class where the REQUIRES_NEW annotated method is. I understand that in this case it was our code that was wrong so it's not really a bug in Jboss way of handling transactions. But it's really strange that there isn't as much as a warning in the logs that there is a race condition. I could restart the Jboss and get a different behavior than before the restart. I think that's strange. I haven't found much about that this could be a problem after googling so I hope that this answer can save some time for someone else.
Edit: It's not about multiple implementing classes, just about declaring all the methods in the interface annotated @Local, otherwise JBoss might ignore the attributes